Recent anomaly detection studies have shown that residual features, which model the discrepancy between normal and abnormal patterns, provide strong generalization to anomalies unseen during training. However, existing residual-based approaches mainly rely on visual discrepancies derived from image representations, while semantic discrepancies that can be explicitly expressed through language remain largely underexplored. To address this problem, we propose TIRAD, Text-guided Image Residual Learning for Generalist Anomaly Detection, a multimodal residual learning framework that defines the discrepancy between textual representations of normal and abnormal states as a Text Residual Feature and uses it to guide the Image Residual Feature. Specifically, a Text Fusion Module converts patch-level text residuals into a gate that modulates the image residual of each patch. The gate retains the visual discrepancies that the text residual identifies as defect-related and attenuates the others, leading to more discriminative anomaly representations. Extensive experiments on industrial benchmarks and industrial-to-medical transfer demonstrate that the proposed method outperforms existing image-centric residual learning methods.
Direct non-binary constituency parsing must identify an ordered sequence of children at each phrase-level reduction. We extend our earlier delimiter-guided bottom-up construction to a common calculus for top-down, in-order, and bottom-up parsing. At reduction, the rightmost delimiter marks the left boundary, the nonterminal marker supplies the mother label, and the completed items in the delimited segment determine the children and arity. We use a stack invariant to prove soundness and completeness and derive the exact length of the static oracle's action sequence. A projection-and-lifting theorem shows that each delimiter-guided system and its canonical counterpart generate the same set of trees under both the structural and grammar-licensed transition relations. In controlled experiments on PTB and CTB, the delimiter-guided parsers achieve similar labeled accuracy to their canonical counterparts across all traversal orders. For bottom-up parsing, delimiter guidance reduces the number of action classes and increases the number of supervised decisions per tree.
Software developers spend significant time reading and understanding code, increasingly supplemented by generative AI. While these tools aim to reduce the difficulty of reading code, interacting with them introduces a dual-task environment that may impose extraneous cognitive load that is challenging to capture. This study investigates how access to generative AI affects developers' real-time cognitive workload, code comprehension performance, and self-perceived code comprehension strategies. We conducted a 3x2 within-subjects study with 20 novice CS students reviewing Java code snippets of varying lengths with or without AI assistance, followed by an AI-unassisted comprehension quiz. Cognitive workload was continuously captured using a Detection-Response Task (tDRT) during both reading and quiz phases, and compared to post-quiz NASA-TLX and SEQ scores. AI assistance significantly increased objective cognitive workload, raising tDRT reaction times by 11.2%, yet participants' perceived workload remained unchanged. Code comprehension scores were high under both conditions, with no evidence that AI assistance improved comprehension at the difficulty levels evaluated; a ceiling effect limited power to detect benefits on harder tasks. Qualitatively, participants reported trusting the AI conditionally -verifying its output against their own reading for more complex snippets, but directly trusting it for simpler code- and, while many felt their reading process was unchanged, a notable subset shifted toward skimming or reading the AI's explanations in place of the code itself. These findings reveal that developers do not accurately perceive the cognitive resources they expend when managing AI tools. This study provides evidence of added cognitive effort associated with using generative AI, pointing towards a need to account for this dual-task penalty in software engineering tools and education.
Dexterous grasping in robotics relies on distinct hand configurations depending on object geometry. We introduce BetaGrasp, a 3D diffusion policy that conditions a DP3-style point-cloud policy on a four-class grasp label (crimp, sloper, pinch, and jug), running on a Franka arm with a 16-DoF LEAP hand. We also introduce a climbing-hold benchmark: a spring testbed whose ratchet gives a slip-force reading in newtons. In our analysis the policy's point-cloud encoder did not separate the four holds, so the scan alone gave it little to choose a grasp from. Without the label (a DP3-style baseline), simulated rollouts from the robot's reset pose all end in a jug grasp, whatever the hold (25% correct grasp type, versus 94% for BetaGrasp). The label works as a switch for which grasp the policy performs. On 80 paired hardware trials, BetaGrasp raises average slip force from 9.6 N to 14.3 N, and grasps slipping fully off the hold drop from 44% of trials to 10%.
Many real-world applications require language models to internalize knowledge that was absent from pretraining, to incorporate continually updated facts, or to learn new domain skills. Self-distillation has emerged as a natural approach for this task. A model conditioned on the source material acts as the teacher, and an unconditioned student is trained to reproduce the teacher's behavior closed-book. We propose Speculative Self-Distillation (SSD), an efficient per-token mixed-policy method for self-distillation. The student generates each rollout by default, while the teacher intervenes only at positions where its next-token distribution substantially diverges from the student's. Unlike off-policy distillation with teacher-generated prefixes, this design keeps training rollouts close to the student's test-time behavior, while avoiding the low-signal states that fully on-policy rollouts drift into. In effect, SSD lets the teacher steer the rollout exactly where the privileged context changes the prediction. On three closed-book QA benchmarks measuring fact recall, knowledge updates, and skill acquisition, SSD nearly matches on-policy distillation while using 38-69% fewer supervised tokens and consistently outperforms off-policy distillation on the held-out test set. On a forgetting stress test, SSD retains the most general capability while also achieving the best in-domain accuracy and using less supervision than the on-policy baseline. Overall, SSD establishes a new accuracy-efficiency-forgetting Pareto frontier for self-distillation.
Large language models and the multi-agent systems built around them could accelerate materials discovery by connecting scientific knowledge with experimental decisions. Existing benchmarks establish capabilities in domain knowledge, multimodal interpretation and scientific tool use. As language models increasingly support materials research workflows, their ability to distinguish closely related material states and reason accurately with quantitative evidence remains insufficiently evaluated. These abilities are essential for comparing material responses, assessing mechanisms and maintaining reliable judgements as experimental evidence accumulates. We introduce MatScope to evaluate three connected capabilities. M1 tests unaided, condition-specific reasoning through 200 tasks spanning ten research functions, using aluminium and high-entropy alloys as representative material systems. M2 tests agent-assisted transfer of related knowledge to unresolved materials questions. MatEvoBench-20 tests initial forecasts across scientific dependency chains and revisions of still-hidden responses as observations accumulate. Across eleven model deployments, including GPT-6 Astra and Claude Fable 5.1, given-input calculations and many physical explanations are handled well. Fine-grained response estimates, comparisons between similar states and quantitatively supported choices remain substantially weaker. Related literature improves selected estimates, while condition and baseline mismatches limit transfer. Progressive observations calibrate some downstream responses and destabilize earlier accurate predictions. MatScope establishes measurable targets for developing reliable scientific reasoning in future self-evolving materials laboratories and materials-aware world models.
Predicting multi-step chemical reaction mechanisms requires identifying plausible intermediates and correctly ordering the elementary reaction steps. Existing computational methods focus on single-step elementary reaction prediction, limiting their ability to generalize to unseen reaction spaces. We introduce a graph-guided framework for multi-step chemical mechanism prediction that leverages a chemical mechanism knowledge graph (ChemMechKG) to guide large language models (LLMs). To improve pathway prediction, we develop a graph-aligned reaction mechanism predictor (GRMP) that uses graph neural networks to generate synthetic candidate pathways for pathway optimization. During inference, we design a relation retrieval (RR) module to extract relevant graph evidence from the KG for pathway refinement. Experiments show that GRMP achieves the highest pathway accuracy among all compared methods on ChemMechQA and oMe-Gold, reaching 48.1% and 57.9%, respectively. On a fixed sample of 110 out-of-distribution reactions from ChemMechQA-Hard, GRMP achieves 36.4% top-1 pathway accuracy. LLM-as-Judge and human assessment further demonstrate its ability to generalize to unseen reactions.
Multi-teacher on-policy distillation (MOPD) merges the capabilities of multiple domain-expert teachers, such as mathematical reasoning, code generation, and instruction following, into a single compact student. Since all domains share a limited training budget, how to allocate it across domains directly determines the quality of distillation. Existing methods allocate the budget by each domain's learning progress, but offer no guarantee for lagging domains. We therefore introduce distributionally robust optimization (DRO) into MOPD to prioritize the worst-performing domain. However, the per-domain training signals are estimated from a few rollouts and are inherently noisy, so directly applying DRO mistakes noise for the worst case, lets occasional spikes hijack the budget, and protects inherently hard domains rather than truly regressing ones. To address these issues, we propose RDS-MOPD, which keeps the DRO objective and redesigns its three parts: it shifts the allocation only when differences clearly exceed the noise, bounds the influence of any single anomaly, and defines the worst case as the regression from each domain's own best level, all with theoretical guarantees. Experiments show that RDS-MOPD outperforms recent competing methods and ranks first on most mathematical reasoning, code generation, and instruction-following benchmarks.
Streaming 3D reconstruction with recurrent states enables long-video scene understanding under bounded memory, but turns reconstruction into a state-evolution problem: each incoming frame updates a compact scene memory on which all subsequent predictions depend, so unreliable updates overwrite historical context and accumulate into long-horizon geometric drift. We analyze CUT3R-style recurrent updates through a direction--magnitude decomposition, in which the magnitude sets the update gain, i.e., how strongly an observation is written into the state, and the direction encodes the geometric intent, i.e., where the update drives the state. This decomposition provides a unified view of CUT3R and its variants, revealing that existing training-free extensions perform magnitude-only control: they can attenuate an unreliable update but cannot correct its direction. We propose HIST3R, a training-free framework built on CUT3R that controls both components using history. HIST3R rectifies the update direction using locally consistent historical states, and modulates the update magnitude according to the visual discrepancy between the current observation and its history, suppressing redundant updates while preserving complementary evidence. Across camera pose estimation, 3D reconstruction, and video depth estimation, HIST3R consistently improves long-horizon performance while retaining the memory footprint and inference speed of CUT3R. With 1000 input frames, it reduces ATE by 23.87% on ScanNet and 33.65% on TUM-Dynamics relative to the strongest competing method; with 500 frames, it improves median reconstruction accuracy over the strongest CUT3R-based variant by 56.52% on 7-Scenes and 33.96% on NRGBD.
Large materials design spaces often contain rare high-performing configurations that are difficult to identify through trial-and-error exploration or purely representative machine-learning (ML) sampling. In this work, we present FruBO, a flexible Bayesian Optimization framework for accelerating the discovery of optimal configurations in computationally demanding materials design problems. As a case study, we investigate multilayer Moir\'e patterns generated by the superposition of six rotated hexagonal lattices, where the objective is to identify the rotational conformation that maximizes pore-size uniformity, quantified using the Jain fairness index. The resulting benchmark contains 600,000 unique configurations and exhibits a highly skewed objective distribution, making the global optimum difficult to locate by random sampling. We compare random exploration, Gaussian Process-based Bayesian Optimization, and Bayesian Optimization using a lightweight CatBoost surrogate with committee-based uncertainty estimation. Random sampling identifies the global optimum only after 533,012 evaluations, requiring approximately 31 hours, while Gaussian Process-based BO becomes impractical due to GPU memory exhaustion after 290 acquired samples. In contrast, FruBO with CatBoost identifies the global optimum after 1,470 acquired samples and approximately 1.15 hours. These results demonstrate that scalable surrogate models can substantially improve the feasibility and efficiency of BO in large nanomaterials design spaces.
Uncertainty quantification in prediction is a central problem in statistics and machine learning. Conformal prediction provides a flexible approach to constructing prediction sets with valid marginal coverage under exchangeability. In this paper, we investigate conformal prediction for both one-step and multi-step forecasting from a single stationary time series, addressing temporal dependence and the additional challenges of joint prediction across multiple horizons. We propose a kernel-based framework, called Time series Residual Analysis for Calibration with Kernels (TRACK), for estimating conditional quantiles of residual scores and constructing prediction sets from rolling-window forecasters. The framework accommodates heterogeneous uncertainty across horizons and accounts for dependence induced by rolling estimation, while allowing predictor memory and forecast horizons to grow with the sample size. Our finite-sample bounds quantify how temporal dependence and overlapping score windows affect calibration and coverage. We further derive bounds for conditional quantile estimation and pointwise conditional coverage discrepancy. The guarantees extend to more general predictors that admit an accurate rolling-window approximation. We also derive matching minimax bounds over classes of known score maps, clarifying how overlapping score windows affect calibration accuracy. Experimental results from simulations and applications to wind energy generation and bike sharing demand show that our methods improve regional conditional coverage while maintaining competitive interval widths.
Protein language models (PLMs) are standard tools for variant-effect prediction, protein annotation and design, but the compute needed to train and serve them keeps growing. We present HyPLM-Flash, an efficient and open PLM trained by Hyra 2.0, the research agent developed by our team. HyPLM-Flash is a 650-million-parameter masked language model trained on 726 billion tokens, 0.73 times the data of ESM-2 650M and a fifth to an eighth of the data of E1-600M and ESM C 600M. It runs on an inference engine with fused kernels and INT8 matrix multiplication, embedding about 130,000 residues per second on a single NVIDIA H20 GPU, 1.9 to 2.7 times faster than other 600M-class encoders. In a matched evaluation against 28 public checkpoints on 32 tasks, HyPLM-Flash reaches a ProteinGym zero-shot Spearman correlation of 0.409, statistically tied with ESM-2 650M, ESM C 600M and E1-600M, and the highest organismal-fitness score among self-supervised single-sequence models. It gives the best long-range contact precision of all models evaluated, including ESM-2 15B and ESM C 6B, and a lower masked-residue loss than ESM-2 650M, ESM C 600M and E1-600M on held-out sequences and on recently released PDB chains. Matching state-of-the-art accuracy at a fraction of the training and inference cost, HyPLM-Flash sets a new Pareto frontier for protein language models. We have released the model weights and the inference engine at https://github.com/Tencent-Hunyuan/hyra-results/tree/main.
Hypergradient (HG) methods for bilevel reinforcement learning (RL) typically assume a white-box follower whose reward is known. In practice, however, the leader rarely knows the follower's reward function and can only observe its trajectories. We address this black-box setting and propose Reward-Estimated Hypergradients (RE-HG), which recovers the follower's reward from these trajectories via inverse RL (IRL). Yet IRL identifies the reward only up to potential-based shaping, a transformation that leaves the follower's behavior unchanged but biases the HG. We prove that, for linear rewards, this bias cancels when IRL estimates are aggregated across sufficiently diverse dynamics, which the leader's own updates naturally supply. To handle finite samples, RE-HG truncates small eigenvalues of the aggregated feature covariance, restoring the identifiable subspace and stabilizing the estimation. Experiments show that RE-HG estimates gradients aligned with a white-box Oracle and achieves comparable performance; empirically, the truncation also suppresses HG explosions, letting RE-HG avoid local optima that trap the Oracle in sharp reward landscapes.
Inverse folding designs amino-acid sequences for a given protein backbone, and screening needs many candidates that are both diverse and well folded. We present HyInv, a 1.96-million-parameter backbone-conditioned graph network with an encoder-decoder architecture, trained by Hyra 2.0, the research agent developed by our team: pretrained with random-order autoregressive sequence modeling, followed by structure-reward reinforcement learning. At inference, a grouped draft-and-rewrite sampler that uses separate draft and rewrite temperatures improves refolding TM-score at matched diversity. An optimized implementation with BF16 numerics designs 128 sequences for a 402-residue backbone in 50.6 ms on a single NVIDIA H20 GPU, about 52 times faster than the reference sampler, and an exact FP32 path that returns the reference sequences is 8.7 times faster. Across five benchmark datasets and seven dataset-predictor combinations, HyInv reaches a higher average scTM than all ten external baselines at comparable or higher sequence diversity, with large margins over ProteinMPNN and ProteinZero V2. The same default setting supports de novo backbones, pocket-constrained libraries, enzyme redesign with the catalytic site kept, and loop redesign. We have released our code and model at https://github.com/Tencent-Hunyuan/Hyra-results/tree/main/AI4Science/hyinv.
In this work, we study whether frontier agent can continually improve real-world AI infrastructure over long horizons. We introduce Infrathon, a benchmark of diverse, long-horizon AI infrastructure optimization tasks across different levels. We evaluate six models across 150 runs with 24-hour budgets, totaling over 10,000 GPU-hours. Our results show that model rankings vary with the evaluation horizon, while 135 of the 150 runs achieve further improvements during the second half of their budgets. These findings indicate that shorter evaluations can overlook both optimization gains and changes in relative model performance. We also find agents can discover solutions that outperform human-designed baselines: Claude Opus 5.5 increases the mean KV-cache compression ratio from 6.82% with the human-designed Quant-VideoGen codec to 19.86%. Analysis of agent trajectories identifies task violations, unreliable testing, and premature completion as barriers to sustained improvement. Together, Our Infrathon framework provide resources for constructing long-horizon tasks, evaluating frontier agents, and discovering good infrastructure solutions.
Small multimodal models can misjudge emotion when words, voice, and visual cues conflict. We present EmoHarness, which runs such a model, the frozen student, inside a harness: the prompts, settings, and verification code around its calls. A human-written starting harness, the scaffold, numbers the student's visual observations, transcript, and prosody descriptions as citable evidence and has the student weigh candidate emotions against it. In Meta-Harness Evolving (MHE), a stronger LLM, the Builder, reads the student's errors and execution traces offline and edits the harness; static checks and validation gates decide which edits are kept, and the Builder never sees evaluation items. On 5,507 MME-Emotion items, EmoHarness raises task-macro Recognition by 7.6-13.4 points over a single video-only call for all four students and both Builders. Transcripts and prosody descriptions supply most of this boost (+5.38 to +10.71); the scaffold changes the audio-augmented call by -0.79 to +2.54, partly through extra multi-label predictions, and evolved edits add at most +0.95, although no evolved increment survives Holm correction. The harness's emotion context also transfers to IntentQA, with small gains for three of four students, none of which survives correction. EmoHarness thus boosts small multimodal students chiefly by supplying and organizing audio-derived evidence.
Talking-head generation models, which synthesize realistic facial animations from audio, are increasingly vulnerable to misuse in multimodal deepfake scenarios. Protecting such systems remains challenging, as talking-head models fundamentally rely on precise audio-visual alignment, while effective audio protection methods remain largely underexplored. In this work, we propose MUTE, an audio protection framework that uncouples the underlying audio-visual alignment through a multi-level strategy. Specifically, MUTE combines (1) Alignment-level Disruption, which directly perturbs cross-modal attention to disrupt structural alignment, and (2) Representation-level Degradation, which perturbs temporal audio embeddings to degrade their representations. To improve robustness, perturbations are constrained in the STFT domain to high-energy regions, making them resistant to post-processing, denoising, and adaptive attacks. An optional speaker-level objective further mitigates potential bypass via TTS-based resynthesis. Extensive experiments demonstrate that MUTE consistently degrades lip synchronization across both white-box and black-box talking-head models while preserving perceptual audio quality. The proposed method remains effective under real-world transformations and can be combined with image-based approaches to provide stronger multimodal protection. The source code and project page are available at https://github.com/jindl465/MUTE.
Multimodal Emotion Recognition in Conversation (MERC) combines lexical, acoustic, visual and conversational cues, but accepting audio and video does not establish their effective or robust use. We audit off-the-shelf Omni models while holding the target transcript and prior textual context fixed and adding, removing or replacing target audio and video. In the primary evaluation of Qwen2.5-Omni-7B, Baichuan-Omni-1.5 and MiniCPM-o 2.6 on MELD and the full five-session six-class IEMOCAP set, T+A improves Accuracy in all six model-dataset pairs, with four gains significant after false discovery rate (FDR) correction. Across three source mappings, unrelated A/V (RandA/V) lowers Accuracy relative to Full in all 18 comparisons; all 18 remain significant under dialogue-cluster inference and BH-FDR correction, both on common-valid pairs and when invalid/OOM outcomes are treated as incorrect. Comparisons with modality-omission baselines further distinguish replacement penalties from mismatch-specific harm. A Qwen/MELD case study finds significantly greater gold-vs-best-candidate margin recovery in Late than Early or Middle under decision-query audio knockout, relative to an edge-count-matched Video-Sham control, using Holm-corrected tests. These results distinguish genuine-modality benefit from sensitivity and vulnerability to misleading input.
Many problems across the sciences compare two stochastic processes through the sequences they generate. Reinforcement learning asks what a new policy would return from the trajectories of an old one; molecular simulation asks how much more a ligand binds from samples of the unbound state; climate attribution asks how a heat wave shifted under warming; and language-model post-training asks what a deployed agent will do from logged interactions. In every case, one quantity would answer all these questions at once: the ratio of the two probability laws over their generated sequences, the trajectory density ratio. This ratio exists only under strong conditions. When one process reaches states the other cannot, the ratio does not exist; when their overlap is thin, its variance grows without control with the sequence length; and when they concentrate on different lower-dimensional sets, the ratio is undefined altogether. Weak support of this kind is the norm for sequence data, not a sampling accident, and its price is paid in identifiability rather than in noise alone. This survey organizes the field around a three-level diagnosis of weak support and a map of the nine retreats that methods take when the full ratio is out of reach, from direct estimation and reweighting through staged decomposition, stationarization, and partial-identification bounds to smoothing, transport couplings, operator spectra, and direct estimation of a single functional. Each retreat keeps part of what the ratio offers and pays for the rest, and each application domain has standardized the retreat its data force. The survey closes with open problems: testable diagnostics at every level, partial identification at the trajectory level, a rate theory of time-amplified error, and benchmarks with ground truth.
Recent Vision-Language-Action (VLA) models have begun incorporating thermal sensing, but existing approaches typically treat thermal images as ordinary visual inputs, providing limited support for temperature-aware relational reasoning in manipulation. To address this limitation, we propose Thermo-VLA with a ThermoRes Encoder, which decomposes thermal images into cold and hot attribute representations and injects them into the RGB token through language-conditioned residual modulation, enabling temperature-driven target selection and continuous manipulation. We further construct six real-world robotic tasks on a dual-arm Unitree G1 platform, spanning basic grounding to relational reasoning. Experiments show that Thermo-VLA achieves an average success rate of 0.86 across four thermal-dependent tasks, outperforming a pi_0.5 baseline that directly incorporates RGB-thermal observations by 18 percentage points. Under unseen-object and scene variations, Thermo-VLA maintains an average success rate of 0.78, achieving the best generalisation performance among the compared methods. Beyond thermal sensing, we transfer the same residual modulation mechanism to colour attributes and observe consistent improvements in both simulation and real-robot experiments, indicating its potential to generalise beyond thermal attributes. Project website: https://j-x-chen.github.io/Thermo-VLA-web/